BOSTON (Reuters) - Many opinion polls show Democrat Hillary Clinton leading Republican Donald Trump in a tight race for the Nov. 8 U.S. presidential election, but any one of four factors may make the outcome harder to predict.
Among the challenges for pollsters: The historic unpopularity of both candidates, the potential Election Day voter response to the polls themselves, the growing abandonment of landlines for cellphones, and the rise of online polling.
Some high-profile stumbles worldwide - including opinion polls that missed Colombia’s Oct. 2 rejection of a peace deal - underscore how technological, social and cultural shifts have made polling more difficult than ever.
An average of polls compiled by RealClearPolitics.com shows former Secretary of State Clinton beating Trump, a businessman, by 5.4 percentage points. Highlighting the difficulties, the range varies from plus-13 for Clinton to a straight tie.
Trump has said the election is rigged against him and this week, in a Reuters interview, he accused media organizations of tilting the polls deliberately, but he has yet to offer any widely accepted evidence to back up these claims.
Voter turnout in the last few presidential elections has been about 60 percent. But given both candidates’ low overall popularity, turnout this year may fall to as low as 52 percent, said Cliff Zukin, a professor emeritus of political science at New Jersey’s Rutgers University and a former president of the American Association for Public Opinion Research.
That makes it hard to guess who might stay home.
“It’s always been difficult to simulate a likely electorate and I think that’s harder to do in 2016,” Zukin said.
A second pitfall is the effect of the polling itself on voters. Sociologists believe polls can weaken projected winners by making their supporters more confident of the outcome and, therefore, less likely to vote.
The percentage of Trump supporters who expect him to win has dropped to 49 percent in a Reuters/Ipsos poll conducted from Oct. 20-24, down from 74 percent from Sept. 16-20. Clinton supporters’ confidence rose at the same time.
If that leads to a higher turnout of Trump supporters than of Clinton supporters, it might affect the election outcome.
Pollsters caution, however, that the effect of polls on the electorate can only go so far.
“If it were showing Clinton up by 2 points, then it’s certainly possible that it would be within the margin of error that Trump might win,” said Douglas Schwartz, director of the Quinnipiac University Poll.
“But if you’re finding that all of these high-quality polls are showing Clinton consistently ahead, then I think you can trust them,” he said.
One of the biggest factors in polling today is the prevalence of cellphones. About half of Americans have only a cellphone and no landline, according to Federal Communications Commission data, more than double the number who were wireless-only in 2010.
This makes it harder and more expensive for pollsters to gather a truly random sample of opinions because U.S. law prohibits computerized auto-dialing (also known as robocalls) to cellphones and there is no central directory for cellphones.
Polling cellphones can costs 30 to 50 percent more than polling land lines due to requirements that the numbers be dialled manually, according to Pew Research Center estimates. This has led polling outfits to generally rely on lower sample sizes to come up with results.
Calls to cellphones are also more easily screened by their users, and as a result pollsters say they connect with just 10 percent of people they try to contact, down from 80 percent a few decades ago.
But Courtney Kennedy, director of survey research at Pew said, “In terms of data, the quality is better.” Many respondents on cellphones are young and racially diverse, she said.
Others, such as Reuters/Ipsos, conduct surveys online. This allows them to reach large numbers of people at lower cost. But because participants are volunteers in many cases, rather than selected at random, segments of the electorate may be left out.
The lower or skewed response in both cellphone and online polls can pose a challenge with pollsters having to adjust results to match the real world. To accomplish this, they weight more heavily the opinions of types of voters under-represented in their surveys.
Pollsters use population statistics, experience and intuition to do this. For instance, if the proportion of men who respond to a survey is lower than their proportion of the overall population, the pollster will adjust the finding to try to even that out.
“Then you’ve built in an assumption that the males that didn’t respond are like the males that responded. And that’s an unknowable fact,” said Robert Groves, provost of Georgetown University, a social statistician and author of seven books on polling who served as director of the U.S. Census Bureau from 2009 through 2012.
But such modelling can work well in some cases.
A study by Columbia University statisticians, published in 2014 in the International Journal of Forecasting, showed that a poll of users of Microsoft Corp’s Xbox gaming system could be used to accurately predict the election’s outcome.
The gamers were far younger, whiter and more male than the U.S. electorate and predicted a sweeping victory for Republican Mitt Romney in 2012. But when the results were weighted according to voters who turned out in 2008, they predicted President Barack Obama’s 2012 re-election.
Getting the weightings right this time could be a tougher job because Trump has attracted many supporters who historically have voted erratically, or not at all.
“We can’t apply 2008 and 2012 models to 2016,” said Ashley Koning, director of the Eagleton Center for Public Interest Polling at Rutgers University.
Reporting by Scott Malone; Additional reporting by Steve Holland in Boynton Beach, Florida, and Maurice Tamman in New York; Editing by Richard Valdmanis and Howard Goller